An Intrusion Detection System (IDS) monitors network traffic continuously, allowing for real-time detection of anomalies and potential threats. This capability to alert administrators to suspicious activities allows for quick responses, reducing potential damage. While numerous studies focus on IDS, the fundamental components of Intrusion Detection Algorithms (IDA), including imbalanced datasets, feature engineering, and model design, have not been thoroughly examined. It is necessary to thoroughly investigate various strategies for handling imbalanced datasets such as resampling methods, the Synthetic Minority Over-sampling Technique (SMOTE), Generative Adversarial Networks (GANs), and several feature engineering techniques such as Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Autoencoders (AEs). Different model designs, both ensemble and non-ensemble methods, should be explored for their applications, challenges, and future research directions. This investigation will be a useful resource for researchers and practitioners advancing IDA technology. This paper presents a model design that employs ensemble learning algorithms integrating SMOTE to address imbalance datasets, unsupervised learning techniques like PCA for feature selection, and a supervised learning embedded framework such as bagging for classification using Random Forest to detect anomaly-based intrusions in IoT networks. The model is evaluated using various datasets, including UNSW-NB15, CICIDS2017, Bot-IoT, and ToN_IoT, for the purpose of multiclass classification. The performance of the model is assessed based on several metrics, including accuracy, F1-score, recall, and precision.